Econometrica, Vol. 66, No.

6 (November, 1998), 1279-1298

THE NOAH'S ARK PROBLEM

BY MARTINL. WEITZMAN

This paperis aboutthe economictheoryof biodiversitypreservation.A cost-effective-

ness methodologyis constructed,whichresultsin a particularformulathat can be used as
a criterionto rankprojects.The rankingcriterionis sufficientlyoperationalto be usefulin
suggestingwhat to look at when determiningactualconservationprioritiesamongendan-
geredspecies.At the same time, the formulais firmlyrootedin a mathematically rigorous
optimizationframework,so that its theoreticalunderpinningsare clear. The underlying
model,called the "Noah'sArk Problem,"is intendedto be a kind of canonicalformthat
hones down to its analyticalessence the problemof best preservingdiversityunder a
limitedbudgetconstraint.
endangeredspecies,diversitytheory.
KEYWORDS:Biodiversity,

INTRODUCTION

THE PRESERVATION is plagued by the absence of a workable

OF BIODIVERSITY
cost-efectiveness framework,withinwhich, at least in principle,basic questions
can be posed and answered.Currentapproachesto endangeredspecies protec-
tion seem almost completely lacking in theoretical underpinningsthat might
reasonablyguide policy.As a result,we do not have rigorouslygroundedcriteria
for choosing among biodiversity-preserving alternatives,and it is difficult to
evaluateperformance.
There are several motivations for this paper. Essentially, I am trying to
introduce an analytical frameworkthat represents a useful way of thinking
about the economicsof diversitythroughthe mediumof an abstractmathemati-
cal model. The "Noah's Ark Problem"is a parable intended to be a kind of
canonicalform of the simplestpossibleway of representinghow best to preserve
biodivers-tyunder a limitedbudget constraint.Throughoutthe paper,emphasis
is laid upon simplicityof structure,in order to get at the analyticalessence of
the problem.
While several themes are developed in this paper, the main motivating
question is how to determine basic priorities for maintainingor increasing
diversity.The centralissue is to developa cost-effectivenessformulaor criterion
that can be used to rank prioritiesamong biodiversity-preserving projects.The
formulashould be operationalenough to be useful in suggestingwhat to look
at when actually determiningconservationpriorities,e.g., among endangered
species. At the same time, the methodology should be firmly rooted in a
mathematicallyrigorousoptimizationmodel, so that the theoreticalfoundation
for the cost-effectivenessrankingcriterionis made very clear.
In talking about biodiversitypreservation,there is always a question about
what is the appropriatelevel of discourse.In principle,the basic unit could be at
1279
1280 MARTIN L. WEITZMAN

the level of the molecule, cell, organ,individual,species, habitat,ecosystem,or

so forth.For the purposesof this paper,I take the underlyingunit of analysisto
be the "species,"althoughI think that the same basic issues and themes of the
paper will arise at any level. In this spirit, the Noah's Ark Problem could be
thoughtof as a generic formulationof "the" core problemof optimalbiodiver-
sity preservationunder a budget constraint.
Suppose, then, that the underlyingconservationunit i essentiallystands for
"species i." In the context of this paper, it is useful to conceptualize a
"conservationproject"as follows. "Projecti" is some preservationaction that
increasesthe probabilityof survivalof species i by APi at a cost of Ci. Let Ub
representthe directutilityof how muchwe like or value the existenceof species
i. (After all, most of us like Pandasmore than mosquitoes.)Withouttryingto be
precise about it at this stage, let the "distinctiveness"of i, meaningroughlyits
differenceor distancefromits closest resemblingunit, be Di. (Imaginesomehow
that Di has been made commensuratewith Ui.) Then the following formula
might appear on heuristicgroundsto be conveyingroughlythe right priorities
for rankingalternatives:

(1) Ri = (Di + U) CL).

As a rankingcriterion,Ri is a measureof the "expectedmarginaldistinctive-

ness plus utility per dollar." Interpreted loosely, when making preservation
decisionsthe conservationauthoritiesare asked to look at the four factors:Di,
Ui, APi, Ci, which seems reasonable,combinedin such a way as describedby
formula(1), which also seems intuitivelyplausible.The proposed formula has
some heuristicappeal as a rule of thumb,and it seems to make sense generally,
but can it be groundedin any theoreticalconstruct?Does the rankingcriterion
(1) have a formaljustification?
The primarytask of the paperis to present a model and a scenariowhere the
use of Ri as a rankingcriterioncan be given a rigorousunderpinning.The main
question I am asking is this: If the "right assumptions"are made, can a
reasonableoptimizingmodel be producedthat providesa rigorousjustification
for using Ri to select conservationpriorities?The short answeris "yes,"and the
underlyingmodel that providesthe "rightassumptions"I call the "Noah'sArk
Problem."
Tryingto answerthe above, rathernarrowlyposed, question about an appro-
priate ranking criterion forces a researcher to confront a number of basic
theoretical and conceptual issues that go to the core of modeling diversity
preservation.My hope is that the basic conceptual frameworkand the set of
ideas introducedhere for thinkingabout the economics of diversitywill prove
useful even beyond the specificapplicationsof this paper.
The model itself will be at a high level of abstraction.My aim is to build the
simplestpossible analyticalstructurethat capturesthe mathematicalessence of
the underlyingproblem.In so doing,I entertainthe standardeconomictheorist's
 NOAH S ARK PROBLEM 1281

hope that basic insightsinto policywhich emerge from the bare-bonescanoni-

cal-formmodel will remain a useful point of departureeven in a more compli-
cated world.

THE SPECIES/LIBRARY" MODEL OF DIVERSITY

A key point of departurefor this paper involvesconceptualizingthe underly-

ing conservationunit-the "species"-as if it were a "library."Concentrating
Qnthe question of how best to preservelibrariesallowsfor a crisp formulation
of the generic problem of optimallyconservingdiversityunder a budget con-
straint.In this section, then, the basic unit is the library.Each individuallibrary
stands,at an appropriatelevel of abstraction,for a particularspecies.
Let the index i = 1,2,..., n denote a particularlibrary.The set of all n
librariesis denoted S.
The value of a libraryconsists of two components:the buildingitself and the
collection of books that it houses. Libraryi is housed in a building that has
some inherent value as a structure-let this utility be denoted Ui. The other,
presumablyprimary,value of library i is its book collection-suppose this
consists of Mi differentbooks. (In the species interpretation,Ui representsthe
direct\utilityof how much we like or value the existence of species i, while the
Mi"books"are very roughlyanalogousto the gene pool of the species itself.)
The book collections in various librariesmay overlap to some degree. The
definitionof the diversityof S, denoted V(S), is the numberof differentbooks in
the overall library collection S. In other words, the diversityfunction V(S)
measuresthe size of the set that consistsof the union of all the differentbooks
in all the librariesof S.
A simple two-libraryexample may help to illustrate the basic idea. Let S
consist of the two libraries 1 and 2. Let the number of books held jointly in
commonbetween libraries1 and 2 be denoted J. Let E1 stand for the number
of books distinctiveor unique to library1, while E2 is the number of books
distinctiveto library2. Then M1 = E1 + J, while M2 = E2 + J.
In the case of two libraries,it is alwayspossible to give a tree representation
of the book structure.The tree, depicted in Figure 1, has a main branch of
length J that dividesinto two sub-branchesof lengths E1 and E2. (One could
tell an evolutionarystory "as if' the two libraries evolved by "descent with
modification"from a common ancestor.)The diversityfunction in this case is
V(S) = J + E1 + E2. Note that diversityhere has the interpretationof being the
total (vertical)branchlength of the correspondingtree.
We defined diversityfor the particularassemblageS of libraries.But essen-
tially the same idea holds for any assemblageof libraries.For any subset of
librariesQ (c S), define the diversityof Q, denoted V(Q), to be the numberof
differentbooks in all the librariesof Q. In other words, the diversityfunction
V(Q) is the size of the set consistingof the union of all books in the librariesof
Q, where Q may representessentiallyany assemblageof libraries.
1282 MARTIN L. WEITZMAN

IJ

El |lE2

1
2
FIGURE 1.

A critical aspect of the underlyingproblem is uncertainty.To continue the

metaphor,a librarycan burn down, with possible loss of the buildingand the
book collection that it houses. Variouspreventivemeasurescan be undertaken
that lower the probabilityof a fire-at a cost. The essence of the conservation
problem is how best to allocate scarce fire prevention resources among the
variouslibraries.
Uncertaintyis introducedas follows. Let Pi be the probabilityof survivalof
libraryi, while 1 - Pi is the probabilityof extinctionof i-e.g., the probability
that library i burns down. For analyticalsimplicity,I assume throughoutthe
paper that all probabilitiesare independent,since this is the easiest case to
analyze.
As previouslynoted, the direct utility of library building i is Ui, \which
representshow much the librarybuildingitself is liked irrespectiveof its book
contents. Without loss of generality,suppose that the coefficient Ui is normal-
ized relativeto the value of each differentbook,whichis set equal to one. Then
the expectedvalue of all the librarybuildingsis

(2) U(P) = EPiLi.

Turningnow to the book collections,the definitionof expected diversityis a

natural extension of the original deterministicconcept to the case of uncer-
tainty. The appropriatestochastic concept here is the probability-weighted
diversity of all subsets. The expected diversityfunction, denoted W(P), is the
average number of different books. More formally,

(3) W(P)- 5?(

QCS jE=Q
r:
Pi)( k E=S\Q -P)VQ

In the two-libraryexample,

(4) W(P) = P1P2 (M1 + M2-J) +P1(l-P2) M1

- -P pl(1 - -0
M2 + (1
Jr
P2(1 PO4 P2
 NOAH S ARK PROBLEM 1283

which can be rewrittenmore simplyas

(5) W(P1, P2) = MlPl + M2P2 -JP1P2

Note that U(P) representsthe expected utility of the librarybuildings,while

W(P) representsthe expectedvalue of the overall collection of books. Of the
two functions, W(P) is by far the more novel or unorthodox,and the really
innovativepart of this paper deals with it.
The expected diversity function W(P) expresses the average number of
dieferentbooks in the collection.Now we mightlike manydifferentbooks per se,
just as we might like manydifferentcolors simplybecause of the more colorful
world their sheer diversitycreates. This would be a kind of aestheticvalue of
diversity.Or, we might want to have differentbooks for the utilitarianreason
that they are a potential source of new future ideas about medicines,foods, or
whatever.This might be called the informationcontent of a book collection.
These two concepts are actually closely related to each other. As the next
section shows, at a sufficientlyhigh level of abstractionthe aestheticvalue of
diversityis essentiallythe same concept as the informationcontentof diversity.

EXPECTED DIVERSITY AS INFORMATION CONTENT

This section develops a strict isomorphismbetween "diversityas aesthetic

value" and "diversityas informationcontent."It will be provedthat when they
are appropriatelymodeled, the two concepts are formallyidentical.
Suppose we take the average number of books W(P) as an appropriate
measure of the aesthetic value of diversity.What should be the appropriate
measureof the informationcontent of diversity?
A "book"in the present model is a metaphorfor a containerof information.
Supposethat a researcheris lookingfor a particularpiece of useful information.
There are three possibilities:the informationmight not exist in any of the
books; or it might be located in just one book; or it might be redundantly
contained in more than one book. Suppose that the probabilitythat any one
book contains this particularpiece of useful informationis ?. A vivid image is
that e standsfor the probabilitythat in any book drawnat randomis contained
the alchemist'ssecret of how to turn lead into gold. In a more realisticscenario,
the "useful information"being sought might be about a particularcure for a
certaindisease or a specificnew food source.
Considerany assemblageof librariesQ. Then the probabilitythat Q contains
useful informationis the probabilitythat at least one book of Q contains the
useful information.Since there are V(Q) different books in Q, this equals
one minus the probabilitythat all V(Q) books fail to contain the useful infor-
mation.Thus, the probabilitythat Q containsuseful informationis
__ . v
,6 H(Q - (1 - Q).
1284 MARTIN L. WEITZMAN

Note that takingthe derivativeof (6) with respectto e and evaluatingat e = 0

yields the relationship

7H(Q;e) = V(Q).

The expression(6) for H(Q; 8) representsthe probabilitythat the determinis-

tic libraryassemblageQ containsuseful information.But underuncertainty,the
libraryassemblagesthemselvesexist only in a probabilisticsense. Libraryi exists
with probabilityPi. The libraryassemblageQ (c S) exists with probability

(8) P (1-Pk)j
jcQ_ -kcS\Q

It is naturalto define formallythe informationcontentI (as a function of P

and e) to be the probabilitythat the overallstochasticlibrarycollectioncontains
useful informationon a particulartopic. Metaphorically,informationcontent is
just the probabilitythat in at least one librarythat has not burned down is
containedat least one book in which is writtenthe alchemist'ssecret of how to
turn lead into gold.
Combining(6) with (8), the informationcontent of the stochastic library
collection is

(9) I ) (l)(kT1 (1-Pk)) [H(Q; )].

QcS jE=Q kcE-S\Q

What is the relationship between diversity and information content? Essen-

tially, the basic result here is that diversityis the first derivativeof information
content with respect to uncertainty.Takingthe derivativeof (9) with respect to
8, using (7) to evaluate it at 8 = 0, and comparing with definition (3) yields the
basic result:

(10) MI(P;e) -W(P)

d8 =o

Since it is readilyverifiedthat I(P; 0) = 0, equation(10) means that W(P) e

is the appropriatefirst order approximationof I(P; 8) in the formalsense tha,t
(11) J(P; 8) = W(P). * + o(82)

where O(e2) standsfor squaredor higher order terms in 8.

The conclusionis that, for small e, maximizingthe expecteddiversityfunction
W(P) is "essentially"the same as maximizinginformationcontent I(P; 8).
We state this result somewhatinformallyas the following:

THEOREM 1: The expected diversityof a set of librarycollections is "essentially"

the same concept as the information content of the same set of librarycollections.
 NOAH S ARK PROBLEM 1285

What do we want to do with expecteddiversity(equals informationcontent)?

Presumablywe want to pick conservationpolicies that maximizethe sum of the
expected diversityfunction of the books W(P) plus the expected utility of the
librarybuildingsU(P), takingaccountof the relevantbudget-likeconstraintson
P. The simplestcanonicalform of a preservationprobleminvolvingoptimizing
under budget constraints,which at the same time conveyssome useful content,
is what I call here the "Noah'sArk Problem."

THE NOAH S ARK PROBLEM

The "Noah's Ark Problem"is intended to be an allegory or parable that

renders a vivid image of the core problem of maximizingdiversityunder a
budgetconstraint.What is treated here is actuallynot the most generalform of
the underlyingmathematicalproblem.Some slight generalizationsare possible,
but they would come at the expense of diluting a crisp version of the basic
paradigm.
Noah knows that a flood is coming. There are n existing species/libraries,
indexed i = 1,2,... , n. Using the same notation as before, the set of all n
species/libraries is denoted S. An Ark is available to help save some
species/libraries. In a world of unlimitedresources,the entire set S might be
saved.,Unfortunately,Noah'sArk has a limitedcapacityof B. In the Bible, B is
given as 300 x 50 x 30 = 450,000cubits3.More generally,B standsfor the total
size of the budgetavailablefor biodiversitypreservation.
In either case, Noah, or society, must face the central problemof choice.A
basic choice question must be answered. Which species/libraries are to be
afforded more protection-and which less-when there are not enough re-
sources aroundto fully protect everything?I present here the simplestway that
I know to model the analyticalessence of this choice problem.
If species/library i is not put on board the Ark, but left unprotected,its
probabilityof survivingthe flood unaidedis Pi. If species/libraryi is boardedon
the Ark, and thereby afforded some protection, its survival probabilityis
enhancedto Pi. Essentially,boardingon the Ark is a metaphorfor investingin a
conservationproject, like habitat protection, that improves survivabilityof a
particularspecies/library.
A particularlygrimversionof the Noah'sArk Problemwouldmake the choice
a matter of life or death, meaning that Pi= 0 and Pi= 1. This specificationis
perhapsclosest to the old testamentversion,so I am takingliterarylicense here
by extending the metaphor to less stark alternatives.The only mathematical
restrictionhere is then

(12) 0<P <?i?1.

There is also a vital "cost"side to the problem.It is naturalto supposethat if

species/library i is boarded,it takes up some space or room on the Ark-let
this space coefficientbe denoted Ci. Since overallspace is limited, the amount
1286 MARTIN L. WEITZMAN

of room that a species occupiesbecomes a criticalfactor in the choice problem.

For the biblical Ark parable, Ci is measured in units of cubits3. In the real
world, Ci would represent the cost of the project that extends an enhanced
measureof protectionto species/library i.
We can now state the central problem. Noah wishes to select survival
probabilitiesthat maximizeexpected diversityplus expected direct utility. The
Noah's Ark mathematicalprogrammingproblemis to select values of {Pi} that
(13) maximize{PJ[W({Pi}) + U({Pi})]
subjectto the n individualprobabilityconstraints
(14) Pt < Pt < Pt Vi,
and subjectto the overallbudget constraint

(15) Ci =B.

Note that, as formulated,the above programmingproblem is continuousin

the probabilityvariable Pi, and thus allows for strictly"interior"values of Pi,
which fulfill the condition
(16) P < Pt < iy
An "interior"value of Pi, one that satisfies (16), correspondsto partial
protection,or "fractionalboarding"of species/library i on the Ark. "Fractional
boarding"might be given a physical interpretationof boarding only some
fraction of a reproductivelyviable populationsize. Suppose that, when not a
single individualof species/library i is boarded,the survivalprobabilitywould
be Pi =P. At the opposite extreme, if population size Ci is boarded, then
suppose the survivalprobabilityis enhancedto Pi = Pi. Finally,assume that for
"in-between"population size boardings a linear interpolation describes the
survivalprobabilities.Under such assumptions,there is a rigorousunderlying
basis for sayingthat a survivalprobabilityof Pi for species/library i comes at a
budget cost of

(17) Pt
ciIP-

I now seek to characterizethe nature of the solution to the Noah's Ark

Problem.The most distinctiveaspectof the mathematicalprogrammingproblem
(13)-(15) is the presence of the unusual "expecteddiversityfunction"W(P) in
the objective.It mightbe thoughtthat the functionW(P) is so unorthodoxthat
it is difficultto say anythinggeneralthat is also interestingabout the solutionto
the problem.Fortunately,it turns out that a quite strikingcharacterizationis
possible.
 NOAH S ARK PROBLEM 1287

The solution of the Noah's Ark Problemis always"extreme"in the following

sense. Noah, or the conservationauthorities that he symbolizes,should be
concentrating all their resources on maximal protection of some selected
species/libraries,even at the expenseof exposingall remainingspecies/libraries
to minimalprotection.
In an optimal policy, the entire budget is spent on a favored subset of
species/libraries that is affordedmaximalprotection.The less favoredcomple-
mentarysubset is sacrificedto a level of minimalprotectionin order to free up
to the extreme all possible scarce budget dollars to go into protecting the
favoredfew.
While a real-worldinterpretationof this result must be properlyqualified,
because it is only as strongas the underlyingassumptions,the followingthought
leaps to mind.Subjectto the restrictionsof the model, there is some implication
that a conservingagency may want to think more in terms of concentrating
limited resources,ratherthan spreadingthem out thinly.
In order to state the above ideas rigorously,we start with the following
definition.

DEFINITION (Extreme Policy Solution): "Almost all" (n - 1 out of n)

species/librariesare either fullyboarded(Pi = 1-) or not boardedat all (Pi = Pi).
At most one species/library j (the "roundoffspecies/library")is "fractionally
boarded"with interiorprobabilityPj, where PJ< Pj< Pj.
The basic result here is the following:

THEOREM 2: The solution of Noah's Ark Problem is an extremepolicy.

PROOF: I present here a concise version,which the reader should be able to

follow even thoughit is compact.Not everysingle aspect is spelled out, because
to do so requires a lot of algebra and notational detail, which the interested
reader should be able to fill in. All the main steps in the underlyinglogic are
provided.
The proof begins with the observationthat conditioningon the existence of
species/libraries 1 and 2 allows the expected diversityfunction W(P) to be
rewrittenin the form:
(18) W(P) = E [K(Q)] [P1P2V(Q u 1 u 2) + Pl(1 -P2)V(Q U 1)
QCS\1\2

+ P2(1 - P1)V(Q U 2) + (1 - P1)(1 - P2)V(Q)]

where K(Q) is definedas the polynomialexpression:

(19) K(Q)- (HP) ( rI\ (' -Pk)) ?0

Q
jCE= kEC
S\Q\1\2
and set notation such as Q U 1 or Q\ 1 are shorthandfor, respectively,Q u {1}
or Q\{1}.
1288 MARTIN L. WEITZMAN

Mechanicallytaking the second mixed partial derivativeof (18) yields, after

some algebraicmanipulation,the formula
d2W
(20) =K(Q)] J(Q)]
81%
P~2 QcS\1\2

where 1(Q) is just shorthandnotation for the followingexpression:

(21) J(Q) [V(Q u 1) + V(Q u 2) - V(Q u 1 u 2) - V(Q)I.

Now the J(Q) formula expressed by (21) actually stands for something.
Lookingcarefullyat the appropriateVenn diagramshould convincethe reader
that J(Q) stands for the numberof books held jointlyby libraries1 and 2 but
not containedin any other librariesof Q. Naturally,J(Q) is nonnegative.
Thus,the expression(20) for the second mixedpartialof W with respectto P1
and P2 is nonpositiveand independentof P1 or P2. It follows that the expected
diversityfunction W(P) is convex in any two of its variables,holding all other
argumentsconstant.
The objectivefunction (13) is therefore convex in any two of its variables,
holdingall other variablesconstant,while the relevantconstraints(14) and (15)
are linear. Because the optimal value of a convex function maximizedover a
convex set is alwaysattained at an extreme point of the convex set, it follows
that, out of anypair of probabilityvariablesin an optimalpolicy,no more than
one wouldbe strictly"interior"in the sense of satisfyingcondition(16). But this
means that at mostone species/libraryj, out of the entire set S, is "fracdonally
boarded"with interiorprobabilityPj,wherePj< Pj < Pj. Q.E.D.

The truly "extreme"nature of this solution can perhapsbest be appreciated

by setting Pi = 0, Pi = 1 for all i. Then a diversity-maximizingpolicy makes each
species (except for at most one) either totally extinct or perfectlysafe.
It is importantto understandthe intuitivelogic that explainswhy an extreme
policy is optimalin the Noah's Ark Problem.Considerthe two-libraryexample
depicted in Figure 1. As before, let El be the numberof books distinctiveto
library1 and E2 the numberof books distinctiveto library2, while the number
of books held jointly in commonbetween libraries1 and 2 is denoted J. Then
the relevantexpressionfor expected diversityin this case is given by (5).
If library1 alone becomes extinct,then El books are lost. If library2 alone
becomes extinct,then E2 books are lost. However-and this is the crucialpoint
-if libraries1 and 2 both become extinct, then an additionalJ books are lost.
With costs linear,Noah would rathershift probabilitiestowardsavingas fully as
possible one of the librariesat the expense of the other in order to "pin down
the line" of the J books in common. The strengthof this effect of increased
returnsto savingeither one of the librariesis measuredby J, which from (5) is
minus the second mixed partial derivativeof expected diversity,corresponding
to equation(20).
 NOAH S ARK PROBLEM 1289

The idea of "pinning down the line of books in common" seems to me to be an

importantconservationprinciple.In the case of costs that are linearin probabil-
ities, it resultsin policiesof extremeconcentrationof conservationresources.As
the incrementalcost of protectionincreasesnonlinearlyin survivalprobability,
this extremenessof an optimalpolicyis diluted,or even reversedfor sufficiently
strong curvature.Nevertheless,the basic principleremains,and may even give
useful insight into the form of an optimal conservationpolicy under some
circumstances.When costs are approximatelylinear in probabilitychanges, as
might be supposed to be the case for sufficientlysmall APi Pi - Pi, then
Theorem2 articulatesa well-definedsense in which a conservingagencyshould
be concentratingscarce resources on some librariesfor the sake of "pinning
down the line of books in common"that might otherwisebe lost from multiple
extinctions.
Theorem 2 of this section gives a strong characterizationof the form of a
solution to the Noah's Ark Problem.The solution form is an extreme policy
where the preservationchanges of some species/libraries are maximallyen-
hanced by their being boarded completely on the Ark, while others are left
completely behind. But this result does not say which species/libraries are
favored by being selected for boarding, nor explain fully why they are so
favored.
As a means towardsthe end of actuallysolvingthe Noah'sArk Problem,there
is a need to understandbetter the propertiesof the expecteddiversityfunction
and also to incorporate into the analysis some measure of uniqueness or
distinctivenessof a species/library. It turns out that these two issues are
intimatelyrelated. There exists a strikingconnection between what might be
called the "marginaldiversity"of a species/libraryand what mightbe called its
"distinctiveness," meaningloosely a distance-basedmeasureof differencefrom
other species/libraries. To bring out this importantrelation, some definitions
need to be made and some more structuremust be put on the problem.The
additionalstructurethat will eventuallybe imposedon species/libraries is that
they are "as if" createdby a specificprocessof descentwith modification,which
will be called the "EvolutionaryLibraryModel."

" DISTINCTIVENESS" IN THE EVOLUTIONARY LIBRARY MODEL

We now wish to define the uniqueness or "distinctiveness"of a library.

Essentially,we identifyhere "distinctiveness"with the fundamentalmathemati-
cal concept of distance.Consequently,there is a need to startwith the relevant
measure of distance in this setting. The definition of pairwisespecies/library
distanceappropriateto the present context is as follows:
DEFINITION: DistanceD(i,j) is the number of different books contained in
libraryi that are not containedin libraryj.
For the importantspecial case where libraries i and j contain the same
number of different books, distances are symmetricbecause D(i,j) = D(, i)
1290 MARTIN L. WEITZMAN

whenever Mi = Mj. It is importantto note, however,that,distances as defined

above are not symmetricin the general case where Mi is not equal to Mi.
Now let Q be any assemblage of libraries. In the present context it is
appropriateto employthe standardmathematicaldefinitionof the distancefrom
a point to a set:
(22) D(i, Q) minD(i, j).

In the case of Q being the null set, distances are normalized so that
D(i, 0) = Mi.
The interpretationof D(i, Q) is that it representsthe distance of library i
from its "nearestneighbor"or "closestrelative"in Q.
It seems natural to define the distinctivenessof i, denoted Dj, to be the
expecteddistanceof i from its nearest neighboror closest relativein S:

(23) Di (1 j)('H (1 -Pk))D(i,Q).

QsS\i ]E Q kE S\Q\i

What is the possible connection between distinctivenessand diversity?To

answerthis questionsharply,one needs to put more structureon the problem.I
now try to impose a "natural"structureon the book collections of the various
species/libraries, which corresponds,at a high level of abstraction,to the
standardparadigmof evolutionaryrelationshipsamong biological species. The
book collections of the various libraries will be modeled "as if' they were
acquiredby a process of "descentwith modification."
Previously,no restrictionswere placed upon the book collections of the
various libraries.Now I want to suppose that the book collections are "as if'
they were acquiredby an evolutionarybranchingprocesswith a corresponding
evolutionarytree structure.The particularbranchingprocess describedhere is
called the evolutionarylibrarymodel,and it is patternedon the classic paradigm
of biologicalspecies evolution.
The "evolutionarylibrarymodel" is a branchingprocess that explains the
existence of the current library assemblage S as a result of three types of
evolutionary/historicalevents.
(i) Each existing library acquires new books at any time by independently
sampling,at its own rate, out of an infinitelylarge pool of differentbooks. Tfie
independentacquisitionof differentnew books by each librarycorrespondsto
the evolutionof genetic traits when species are reproductivelyisolated with no
gene pool mixingby lateral transfer.
(ii) New librariescan be createdby a "speciationevent."A new branchlibrary
can be founded by adopting a complete copy of the current collection of an
existinglibrary.Henceforththis new librarywill become reproductivelyisolated
and acquireits books independently,as describedby (i) above.
(iii) Libraries can go extinct. When a library is extinguished, its entire
collection of books is lost. Librariesthat have alreadygone extinct in the past
 NOAH S ARK PROBLEM 1291

correspondto lost stem taxa, and do not show up in the set S of currently
existinglibraries.
Figure 2 illustrates the evolutionarylibrary model for the case of three
species/libraries.The firsttwo librariescorrespondto the depictionof Figure 1,
shown in a previous section of the paper. As was pointed out there, any two
librariescan be given an evolutionarytree representation.But for three or more
libraries,the evolutionarylibrarymodel mustbe assumedin orderto have a tree
representation.Hence, three librariesis the simplestcase to analyzewhere the
evolutionarylibrary model is actually imposing additional structure on the
librarycollections.
The evolutionarylibrarystorytold by Figure2 is somethinglike the following.
At the beginningthere was one prototypelibrary,which acquireda collectionof
G different books. Then occurred a "speciationevent." Two "reproductively
isolated"librarieswere created,each startingoff with identicalcopies of the G
different books. Both librariesthen began independentlyto acquire different
books. One of the two libraries eventuallybecame, after acquiringindepen-
dentlyE3 differentbooks, the currentlyexistinglibrary3. The other librarywas
the ancestorof libraries(1, 2). This ancestorlibraryof (1, 2) acquiredindepen-
dently J differentbooks before another "speciationevent"occurred-resulting
in the fission-likefoundingof two identicallibraries.One of these libraries,after
furtherindependentacquisitionof E1 differentbooks,became eventuallylibrary
1. The second library,after furtheracquiringindependentlyE2 differentbooks,
evolvedinto library2. In this way did the currentcollection of libraries{1, 2, 3}
come into being.
I presenthere the "evolutionarylibrarymodel"in a particularlysharpform to
emphasize its essential characteristics.As such, the concept is intended to
representan "ultimateabstraction"of how entities are createdby an evolution-
ary branchingprocess. While the model could tolerate some generalizations,

|G

1 1 l ~~~~~E3
El E2

1 1 ~~~~~~~3
2
FIGURE 2.
1292 MARTIN L. WEITZMAN

such as types of nonindependentsampling,I do not feel it is worth the loss of

sharpfocus to pursuehere the most general possible formulation.
In thinking about the evolutionarylibrary model, it is almost as if the
fundamentalunits are definedas entities that satisfy the three basic axioms.
Since the model is at such a high level of generalization,in a way it does not
really matter whether "evolutionarylibraries"stand for biological species, or
languages,or somethingelse. An "evolutionarylibrary"here is an abstraction
standingfor an entity that evolves as a unit independentlyof other such units,
and which came into being originallyby splittingoff from anothersuch entity. I
think there is some merit in first definingcarefullythe mathematicalessence of
an evolutionarybranchingprocess-i.e., the "evolutionarylibrarymodel"-and
then, in the biological context, defining species to be units that satisfy, not
perfectlybut tolerablywell in practice,such a model.
The evolutionarylibrarymodel naturallygeneratesa correspondingevolution-
ary tree. And when a tree structureis present, it seems to induce naturallyin
the human mind a way of visualizingand comprehendingintuitivelyrelation-
ships amongobjectsthat are quite subtle or complicatedto describewithoutthe
tree. "Tree thinking"representsa prime example of how one picture may be
worth a thousandwords.
Take, for example,diversity,which is definedas the total numberof different
books.In the evolutionarylibrarymodel, diversityhas a naturalinterpretationas
the total (vertical)branchlength of the correspondingtree. For the three-library
example of Figure 2, it is readily confirmedthat the diversityof the three
librariesis the total branchlength G + J + E1 + E2 + E3.
How is "distinctiveness"representedin the contextof an evolutionarySbranch-
ing model?Absent uncertainty,the distinctivenessof a libraryis just its distance
from its nearest neighbor or closest relative. In the tree correspondingto an
evolutionarybranchingmodel, the distinctivenessof a libraryis representedby
its (vertical)branchlength off the rest of the evolutionarytree. Thus, in Figure
2, one readilyconfirmsthat the distinctivenessof libraryi is representedby its
(vertical)branchlength Ei for all i = 1,2, 3.
Note here an importantgeometricrelationship.For an evolutionarylibraryin
the deterministiccase, the loss of diversityfrom extinctionis just the length of
the (vertical)librarybranchfrom the rest of the evolutionarytree. The precise
mathematicalstatementis
(24) V(Q u i)-V(Q) = D(i, Q).
A vividimage is that the extinctionof a librarycorrespondsto snappingoff its
branchfrom the rest of the evolutionarytree. In Figure2, the loss of diversityif
library1 goes extinct is E1. The loss of diversityif library2 goes extinct is E2.
(The loss of diversityif libraries1 and 2 both go extinctis E1 + E2 + J, which is
the total (vertical)branch loss of sequentiallysnappingoff 1 first and then 2
next, or vice versa.)
In the deterministiccase, the loss of diversityfrom extinction of a library
exactlyequalsthe distinctivenessof the library.Essentially,changeof diversityis
 NOAH'S ARK PROBLEM 1293

distinctiveness.This importantrelation generalizes from a deterministicto a

probabilisticsituation.The basic idea can be stated formallyas the following
result:

THEOREM 3: In the case of evolutionarylibraries,

8W
(25) = Di for all Pi E [O,1].

PROOF: Rewritethe diversityfunction(3) as

(26) W(P)PiE= (H I P) ( H (1 -Pk))V(Q Ui)

Qc S\i jEQ k c- S\Q\i

+ (1 -Pi) E ( n Pj) fl ( -P)VQ

Q sS\i i E-Q k c- S\Q\i

Takingthe derivativeof (26) with respect to Pi and collectingtermsyields

8W
(27) = HPi H
f (1-Pk) u(Q i)- V(Q)].
8Pi QcS\i JEQ kcS\Q\i

Combining(27) with (24) and (23) yields the result(25). Q.E.D.

Theorem 3 is a statementthat marginaldiversityand distinctivenessare the

same concept in the context of an evolutionarylibrarymodel. The theorem
representsan appropriategeneralizationto the uncertaincase of the essential
idea from the deterministiccase that loss of diversitycan be visualized as a
branch-snappingevent.
There is, however, a new wrinkle added by the presence of uncertainty.
Distinctivenessnow is associatedwith expecteddistancefrom a nearestneighbor
or closest relative, loosely speaking.Under uncertainty,seemingly symmetric
librariescan have different degrees of distinctiveness,with consequences for
conservationpolicy.It is importantto understandhow this comes about.
To see this effect most starkly,return to the simple two-libraryexample of
Figure 1. Make the tree picturesymmetricby supposingthat both librarieshave
the same number of books so that M1 = M2 = M, and also E1 = E2 = E. Taking
the derivativeof expression(5), and usingrelation(25),yields that D1 = M - JP2,
while D2 = M - JP1.Thus, in a seeminglysymmetricsituation,the more endan-
gered species is less distinctive.
To expose sharplythe underlyinglogic behind this seeminglycounterintuitive
result,supposean extremeexamplewhere P1 = .99,while P2 = .01. Then library
1 is practicallysafe, while library2 is practicallyextinct. For such a situation,
library2 is much less distinctivethan library 1 in the following sense. The
presence of library2 is practicallycontributingonly E different books, since
library1 is almost sure to surviveby itself. On the other hand, the presence of
1294 MARTIN L. WEITZMAN

library1 is practicallycontributingE + J differentbooks, because library2 will

almost surely become extinct. In this sense library1 is much more distinctive
than library2. The principleof "pinningdown the line" of jointly held books
manifestsitself in this exampleby indicatingthe relativelysafe libraryas more
distinctive than the relatively endangered library. Pushing even further the
extremesymmetryassumptions,if the underlyingcosts of changingprobabilities
are the same, then it is optimal to make the safe libraryeven safer at the
expense of makingthe endangeredlibraryeven more endangered-because the
safe libraryis more distinctiveand has greatermarginaldiversity.
At this point all of the necessaryanalyticalapparatushas been developed,and
the paperis readyto begin its main theme of developinga rankingcriterionthat
solves the Noah's Ark Problem.

THE NOAH'S ARK MYOPIC RANKING CRITERION

Suppose Noah wishes to actually solve problem (13)-(15). He wants to

maximizeexpecteddiversityplus directutilitysubjectto the relevantconstraints.
He does not want,however,to mess aroundwith a complicatedalgorithm.Noah
is a practicaloutdoorsman. He needs robustnessand ruggedperformance"in
the field."As he stands at the door of the ark, Noah desires to use a simple
priorityrankinglist from which he can check off one species at a time for
boarding.Noah wishes to have a robust rule in the form of a basic ordinal
rankingsystem so that he can board first species #1, then species #2, then
species #3, and so forth, until he runs out of space on the ark, whereuponhe
battens down the hatches and casts off.
Canwe help Noah? Is the conceptof an ordinalrankingsystemsensible?Can
there exist such a simple myopicboardingrule, which correctlyprioritizeseach
species independentof the budget size? And if so, what is the actual formula
that determinesNoah's rankinglist for achievingan optimalark-fullof species?
The answerto these questionsis essentiallypositive,and along the following
lines.
Providedthat

(28) 'AP- Pi

is "relativelysmall"' (for all i) in the usual sense of the prototypicalsmiall

project justifyingcost-benefit investmentmethodologylocally, then a priority
rankingbased on the criterion

(29) Ri= [Di+ Ui] Ci

Ci

Note that the presumed smallness of AP goes somewhat against the spirit of the biblical version
of Noah's Ark, for which a fair interpretation might be AP= 1. If so, I plead literary license to
justify using the extended metaphor here, because it is so pretty.
 NOAH'S ARK PROBLEM 1295

is justifiedin the sense of givingan arbitrarilyclose first order approximationto

an optimalpolicy.
To intuit why this mightbe so, ask the followingquestion.If we have enough
moneyto adjustprobabilitiesa little bit in a particulardirection,whichdirection
wouldyou choose? This is askingabout the "gradient"of the objectivefunction.
Theorem 4 says that the gradient indicates a derivative of Di + Ui in the
directionof Pi, which impliesa policyof the specifiedextremeform that pushes
probabilitiesto their maximumor minimumvalue.
More formally,we have the followingtheorem.

'THEOREM 4: Maintain the evolutionary-library hypothesis. Suppose one selects a

solution of the Noah's Ark Problem to be of the following form: Thereexists a cutoff
value R* such that

Ri > R* Pi = P (species i is boarded),

Ri < R* Pi = Pi (species i is not boarded).

Then the errorintroduced by this proposed solution is of second or higher order in

{zlPi}.

PROOF: The proof presentedhere is concise.Not everysingle aspect is spelled

out, since to do so would requirea lot more algebraand notation.While some
of the messier details about the applicationof Taylor-seriesapproximationsare
left to the reader,all of the main steps in the underlyinglogic are providedhere.
For all {Pi} satisfying(14),
n E3W
(31) W({Pi}) W({Pi}) + dpi -(Pi-P)
i=1 L ~i
P=pi

where the symbol = stands for a first order approximation,which is arbitrarily

accuratefor sufficientlysmall {APi} in the traditionalsense of a Taylor-series
expansionthat omits only terms of order (Ai)2 or higher.
But from Theorem3,

rSW]
(32) =DPi

Using (32), rewrite(31) as

n
= +
(33) WQ{Pi}) W({Pi}) 5?Di(Pi-P).i
i=l

Now substitutethe linearizedexpression(33) into the objectivefunction(13),

for which it is an arbitrarilyclose approximation.The reduced form of the
1296 MARTIN L. WEITZMAN

linearizedversionof (13)-(15) is now a programmingproblemthat selects values

of {Pi}that
(34) maximizepi,E aicPi+ constant
subjectto the n individualprobabilityconstraints
(35) Pi < Pi < Pi,I li,
and subjectto the overallbudget constraint
(36) E 83iPi= 'y
In the linearizedproblem(34)-(36), the followingdefinitionsare employed:
(37) ai-Di +U
and

(38) 8i- C'

APi
and
C.P.
APi
Using a Taylor-seriesexpansionargument,a straightforward but notationally
very messy lemma shows that the error in the objectivefunction introducedby
using the solution of (34)-(36) as an approximation to the solution of (13)-(16)
is of second or higher order in {APi}.
Now the problem(34)-(36) is in the form of a classicallinear programming
budgetingproblem.The relevantsolutionconcept is to rank"investmentoppor-
tunities"by the ratio criterion:

(40) {a}
and alwaysto favorhigher-rankedprojectsover those of lower rank.
When applied to the Noah's Ark context, and makinguse of (37), (38), tJ;ie
ratio criterion(40) is equivalentto followingpreciselythe statementof Theorem
4. This concludesthe concise proof of Theorem4. Q.E.D.

Theorem 4 representsa culminationof the researchstrategymotivatingthis

paper. It has been shown that a methodologythat "feels" like a traditional
cost-effectivenessapproachcan be constructedto deal with the conservationof
diversity.The myopicrankingcriteriondevelopedhere is sufficientlyoperational
to be at least useful in suggestingwhat to look at when determiningconserva-
tion priorities.At the same time, the formula is rigorouslyderived from an
optimizationframework,so that its theoreticalfoundationsare clear.
 NOAH S ARK PROBLEM 1297

DISCUSSION

The rankingformula(29) encouragesthe conservationauthoritiesto focus on

four fundamentalingredientswhen choosingpriorities:
Di = distinctiveness of i = how unique or different is i;
Ui = direct utility of i = how much we like or value i per se;
APi = by how much can the survivabilityof i actually be improved;
Ci = how much does it cost to improvethe survivabilityof i.
I am not intendinghere to argue that it is easy in practice to quantifythe
above four variablesand combinethem routinelyinto the rankingformula(29)
that defines Ri. The real world is more than a match for any model. The
essential worth of this kind of research is to suggest a frameworkor way of
thinkingabout biodiversitypreservation,and to indicatehow it mightbe backed
by a rigorousunderlyingformulation.
The basic hope is that the formulafor Ri could still be used as a roughguide
or rule of thumbfor decidingconservationprioritieseven in situationswherewe
cannot know Ci, APi, Di, or Ui with any precision. Perhaps one could come
away with a sense that when making conservationdecisions in the name of
preservingdiversity,it might seem like a "good idea" at least to consider the
four factors Di, Ui, APi, and Ci-especially in a policy world so otherwise
lackingjustifiableguidelinesfor endangeredspecies protection.One is perhaps
further encouraged to think that combining these four ingredients into an
overallindex Ri, more or less as indicatedby (29), also seems like a "goodidea"
because it is intuitivelyplausibleand backed by a rigoroustheory in a special,
but not unreasonable,case.
The ultimate hope is that the metaphor of the Noah's Ark Problem, the
associated Myopic Ranking Criterion,the underlyingLibraryModel, and the
rest of the conceptual apparatushave "stayingpower" as a useful guide to
organizedthinkingabout the economicsof biodiversitypreservation.

Dept. of Economics, Harvard University,Cambridge,MA 02138, U.S.A.

ManuscriptreceivedJuly, 1996; final revision receivedJune, 1997.

REFERENCES

BROWN, GARDNER M., AND JOHN H. GOLDSTEIN (1984): A Model for Valuing Endangered Species,"
Journal of EnvironmentalEconomics and Management, 11, 303-309.
CROZIER, Ross H. (1992): "Genetic Diversity and the Agony of Choice," Biological Conservation,61,
11-15.
METRICK, ANDREW, AND MARTIN L. WEITZMAN (1994): "Patterns of Behavior in Endangered Species
Preservation," Land Economics, 72, 1-16.
REID, W. V., AND K. R. MILLER (1989): KeepingOptionsAlive: The ScientificBasisfor Conserving
Biodiversity.Washington D.C.: World Resources Institute.
1298 MARTIN L. WEITZMAN

SOLOW,ANDREWR., STEPHAN POLASKY, ANDJAMESM. BROADUS(1993): "On the Measurement of

Biological Diversity," Journal of EnvironmentalEconomics and Management, 24, 60-68.
(1993): "Searching for Uncertain Benefits and the Conservation of Biological Diversity,"
Environmental and Resource Economics, 3, 171-181.
WEITZMAN, MARTINL. (1992): "On Diversity," QuarterlyJournal of Economics, 107, 363-406.
(1993): "What to Preserve? An Application of Diversity Theory to Crane Conservation,"
QuarterlyJournal of Economics, 108, 157-183.